Labeling Errors in Football Analytics: A Lesson from a Mexican Retirement Event

In my analysis log, there was an entry marked "football." I opened it one evening in Lyon. Inside, no team, no player, not a single minute of football. It was a report on the Feria de Afores 2026, an information fair about retirement funds held from October 8 to 12, 2026 in the Iztacalco district of Mexico City. The event is organized under Consar — the authority overseeing the Sistema de Ahorro para el Retiro (SAR) — with the participation of Afore pension-fund administrators. Attendees are advised to travel by Metrobús, to check in advance whether the procedure they need is actually among those being processed, and to follow official announcements from Consar. No entry fee was recorded. What made me stop was not the content. What made me stop was the label. The label says "football." The content says "pensions." I tell this story because it touches the very place my profession tends to hide: the upstream end of the data. In 2026, when I was 18, I sat at the Balmont ground watching Lyon Duchère host Jura Sud in the CFA — France's fourth tier. Every match report that day mentioned only the scorer of the brace. I noticed a 20-year-old central midfielder: Mohamed Sarr touched the ball 58 times, completed 51 of 55 passes, made 6 interceptions, scored no goals, assisted none. I spent two weeks rewatching four matches, counting every pass by hand, and found that 80% of Duchère's dangerous attacking moves ran through his feet. Balmont does not produce stars; it only reveals who is willing to run more to shine. In 2026, I analyzed 22 matches at the World Cup in Russia, focusing on the French national team. I did not let the 4-2 win over Argentina carry me away. I logged 14 French pressing sequences in the first half and wrote my longest piece on the Pogba – Kanté – Matuidi trio. The 2026 World Cup taught me that the midfield does not need a hero; it needs a rhythm-keeper. The article drew 5,400 reads, but I still waited 48 hours to recheck every figure. In 2026, when European leagues paused for the pandemic, I coded 120 matches across six leagues — Ligue 1, Premier League, La Liga, Bundesliga, Serie A, Eredivisie — against twelve structural criteria: distance between lines, pressing direction, defensive angles. The pandemic season did not destroy football; it stripped away the illusion of attack to expose the pressing framework. That criteria set became a master's thesis in Sports Management and opened the door to my current job. So I understand the value of clean data. And I understand the price of a dirty data line. A football data pipeline runs through three layers. The collection layer: news, reports, statistics, footage. The classification layer: labeling by topic, by league, by event type. The analysis layer: building models, computing metrics, drawing conclusions. When the classification layer is wrong, the analysis layer does not know it is wrong. It keeps running, keeps producing numbers. They are just meaningless numbers. In this case, a pension report was labeled "football." If I were lazy, I would try to analyze its tactics. I would look for a team playing three at the back, for anyone pressing high, for a movement block worth discussing. I would find nothing, because there is nothing to find. The result would be a report full of blanks. Worse, if I wanted a fit, I would invent conclusions to fill the blanks — turning a labeling error into a professional mistake. I trust the pressing map more than the post-match quote. But a pressing map is only correct when the input data genuinely belongs to football. A heat map drawn from junk data still looks real, still smooth, still shows red exactly where the crowd is. That is why I hold that the heat map has become the new astrology: it is so visually persuasive that people forget to ask where the data came from and who labeled it. The same error, placed inside football, looks more familiar than we think. A match mislabeled a "derby," so it enters an intensity sample. A player labeled a "holding midfielder" when in reality he plays right-of-center and his main job is to stretch the flank. A youth league mixed into a professional sample, so fitness data gets compared at the wrong level. Each small error, compounded across rounds, skews every metric behind it. In modern football, space does not appear on its own; it is forced open by the movement block. But before talking about the movement block, I need to be sure I am watching the right match. That is the first condition, and also the least-mentioned one. Why is it least mentioned? Because classification is quiet work. It does not go on camera, does not spark debate, wins no praise. People praise models, charts, correct predictions. No one praises a correct label. Yet that quiet layer is what determines whether the loud layer is real. I have seen this temptation in transfer work. Transfers are not a race for money; they are a race to find the right person for the right gap. With data, the "right person" must first be the "right place." A pension report does not belong in the football stream, no matter what the label says. We usually worry about tactical blind spots — a coach who fails to see the opponent switch formation, a defender who fails to see a striker running behind. Those are on-pitch mistakes, momentary, and visible when the tape is replayed. The real blind spot sits where no one replays: the labeling layer. A mis-topic article makes no noise. It sits quietly in the dataset, waiting to be counted into some sample. The analyst downstream sees only data, not that it was ever filed in the wrong place. The paradox is that the more automated we get, the bigger this blind spot grows. A machine labels faster than a human, but a machine only re-labels what it has already seen. It learns from old data. If the old data carries an error rate, that rate multiplies with every loop. One pension fair slips in once, and tomorrow dozens of other items slip in too, simply because they "look alike." I choose a different response: when the label says "football" but the content holds no football, I do not force out a football article. I flag it as a wrongly sourced item and drop it. This is less glamorous. But if I turn a labeling error into an analysis piece, I do not merely err once — I teach the system that the error is acceptable. The lesson is neither in Mexico nor in football. It is in the order of the work: check the topic before analyzing, verify the label before concluding. Next match, when I open a data file, the first thing I will do is ask myself: is this really football, and who labeled it. If the answer is still vague, I stop — even if the file looks very pretty.